From the Gamboa field season in 2021-2022, we collared 2 full groups of white-nosed coatis; Galaxy and Trago, with GPS and audio collars. In Galaxy group, we discovered that they exhibited fission fusion behaviour. In this report, we will explore the properties of these sub-grouping patterns.

Contents:

1: How to define sub-groups

To investigate the fission-fusion events in Galaxy group, we need to define when the group has fissioned. Data used for this analysis were from the GPS points every 10 minutes from 6am to 6pm from the 25th December 2021 till the 13th of January 2022. For each timestamp used in the analysis, all group members must have had a GPS fix.

Plot 1: Sub-group count with varing radii

We used dbscan: Density-based Spatial Clustering with varying epsilon neighbourhood values For each individual, we made a radius of a set distance (10m to 100m), if another individual was in that radius, they were clustered together.

As you can see from the plots, once the epsilon neighbourhood value reached 50m between individuals, the majority of sub-groups were between 1 to 3 groups. This gives us a good approximation of where the distance threshold should be placed to capture when the group was split. We wanted to investigate this further by looking at the proportion of individuals in each sub-group.

Visualisation of the fission-fusion dynamics every 10 minutes over 12 hours:

2: Proportion of individuals in each sub-group

We used the 50m radius as the cut-off distance to define the sub-groups. The data was split to when the group was divided into 2 and 3 sub-groups accordingly. We then looked at the number of individuals in each sub-group, to understand how these sub-groups arrange themselves when the group has fissioned.

Plot 2: Group split with 2 sub-groups

When the group was split into 2 sub-groups, the majority of splits were either even with 5 individuals in one group and 6 in the other, or the split was from one individual outside of the group, whilst the rest of the group were together.

Plot 3: Group split with 3 sub-groups

When the group was split into 3 sub-groups, the group sizes were either 5, 5, 1, or 9, 1, 1. Therefore, we can show that when the group splits, they either split evenly or the majority of the group is together with 1 or 2 individuals away from the group. The tables below give the counts of individuals in the different split compositions.

Counts for the number of individuals in each sub-group:

4 sub-groups

## 
## 1_1_1_8 1_1_2_7 1_1_4_5 1_2_3_5 1_2_4_4 
##       8       2      14       5       1

3 sub-groups

## 
## 1_1_9 1_2_8 1_3_7 1_4_6 1_5_5 2_3_6 2_4_5 
##    96     4     3    13   125     1     8

2 sub-groups

## 
## 1_10  2_9  3_8  4_7  5_6 
##   78   12   17   16  216

The issue with these results is that single individuals are not technically a sub-group. So I have removed those instances from these data.

Plot 4: Group split with 2 sub-groups omitting solitary individuals

Plot 5: Group split with 3 sub-groups omitting solitary individuals

Here we can clearly see that the majority of split events, the group splits evenly.

3: Sub-group compositions

For data where all individuals had a GPS fix, we calculated the fraction of time every dyad were in the same sub-group. The aim for this was to determine whether individuals have distinct sub-group memberships.

Plot 6: Sub-group compositions

It is clear that coatis in Galaxy group spend more time with certain individuals, so it could be speculated that they have preferences for who they remain with when the group splits. We can also see that one of the sub-groups contains all adult females and the only juvenile of the group, whereas the other sub-group contains 2 adult females and 3 sub-adults. Gus, the adult male spends slightly more time with the adult group, but very little time with Estrella.

4: Within-group compositions

Is there multi-level sub-grouping patterns?

We did the same analysis for when all group members were together to determine whether there is within-group spatial clustering. To do this, we changed the epsilon neighbourhood value (radius) to 10m. So if the group is spread 30m, we would be able to identify which individuals are closer together.

Plot 7: Gus in the group with 10m radius

This matrix shows that there may be some preferences within the full group. Quasar tends to be closer to Pluto, Luna, and Gus. Luna also tends to be closer to Pluto and Gus.

As Gus was not always present in the group, we removed him from the analysis which gave us an additional 42 data points. However, the proximity values don’t change much.

Plot 8: Without Gus in the group with 10m radius

Plot 9: Without Gus in the group with 3m radius

Here, I changed the radius to 3m to look at whether the distance threshold set may affect the results:

Without Gus in the group, we see Quasar closer to Luna and Pluto as well as Saturno. However, with short radii values, the proportion of time individuals are in the same within group sub-group is largely reduced. So these results may be inaccurate interpretations. These plots show that the value we set as the radii is important for how we interpret the social relationships within the group.

Further questions:

With the 1Hz data: